Moving Object Recognition using Stopping human tracking and MRF in Frame Difference

Abstract

In this paper, the authors describe on a moving object tracking method using frame difference and time-space MRF. In case of using the frame difference, generally, it is impossible to track a stopping object because the object is undetectable. In the method, after combination of moving regions of input image and previous image, tracking procedure of stopping region that is continuity of information on stopping region or resumption as a moving region is performed. Concerning an overlapping region occurred by passing each other, boundary region is classified by time-space MRF. By these procedures, it is possible to track moving objects and stopping objects individually. In the paper, the method is explained in detail and an experimental result of cars and pedestrians at an intersection is shown.